Journal of Catalysis 345 (2017) 39–52

Contents lists available at ScienceDirect

Journal of Catalysis

journal homepage: www.elsevier .com/locate / jcat

Efficient volcano-type dependence of photocatalytic CO2 conversion intomethane using hydrogen at reaction pressures up to 0.80 MPa

http://dx.doi.org/10.1016/j.jcat.2016.10.0240021-9517/� 2016 Elsevier Inc. All rights reserved.

⇑ Corresponding author.E-mail address: yizumi@faculty.chiba-u.jp (Y. Izumi).

1 Current address: Innovation Research Center of Fuel Cells, University of Electro-Communications, 1-5-1 Chofugaoka, Chofu, Tokyo 182-8585, Japan.

Shogo Kawamura a, Hongwei Zhang a, Masayuki Tamba b, Takashi Kojima b, Masaya Miyano a,Yusuke Yoshida a,1, Mao Yoshiba a, Yasuo Izumi a,⇑aDepartment of Chemistry, Graduate School of Science, Chiba University, Yayoi 1-33, Inage-ku, Chiba 263-8522, JapanbDepartment of Applied Chemistry and Biotechnology, Graduate School of Engineering, Chiba University, Yayoi 1-33, Inage-ku, Chiba 263-8522, Japan

a r t i c l e i n f o

Article history:Received 30 August 2016Revised 20 October 2016Accepted 22 October 2016

Keywords:CO2 conversionPressure dependencePd/TiO2

Electron trapOxygen vacancyInterface siteSurface plasmon resonanceContour plot

a b s t r a c t

Photocatalytic conversion of CO2 into fuels could mitigate global warming and energy shortage simulta-neously. In this study, the reaction pressure was optimized for CO2 reduction by H2. The major productswere methane, CO, and methanol, and the observed catalytic activity order was Cu or Pd on TiO2 � Ag/ZrO2 � g-C3N4 > Ag/Zn3Ga-layered double hydroxide � BiOCl. Hot/excited electrons due to surface plas-mon resonance could be transferred to CO2-derived species and the remaining positive charge could com-bine with excited electrons in the semiconductor. As the levels of hot/excited electrons became morenegative, the catalysts became more active, except for Ag/ZrO2 and Ag/Zn3Ga-LDH, probably due to lowercharge separation efficiency for intrinsic semiconductors or hydroxides. The reaction order was con-trolled by the partial pressure of H2, demonstrating preferable adsorption of H on Pd. The photoconver-sion of CO2 into methane was optimum at PH2 = 0.28 MPa and PCO2 = 0.12 MPa, but the rates graduallydropped at higher partial pressures due to adsorption of CO2 being hindered by H.

� 2016 Elsevier Inc. All rights reserved.

1. Introduction

Photocatalytic conversion of CO2 into fuels is one of the routesto carbon-neutral fuels, avoiding the net increase in atmosphericCO2 concentrations associated with fossil fuel-derived alternatives[1–3]. TiO2 has been investigated most intensively for this purposeand is known to produce methane and/or CO. CuO supported onTiO2 photoreduced CO2 into methanol [2,3] or methane [1–3].However, when isotope-labeled 13CO2 was used, C-containingimpurities at the TiO2 surface were likely to be reduced to CO[3,4]. The product using Pd/TiO2 was carefully confirmed to be13CH4 [3,5].

We recently reported CO2 photoconversion into methanol usingH2 and layered double hydroxides (LDHs) [6–8]. Industrially, H2 isproduced via cracking and reforming fossil fuels. H2 can instead beproduced sustainably from water utilizing natural light, in a waysimilar to Photosystem II,

2H2O ! O2 þ 4Hþ þ 4e�; ð1Þ

followed by the production of nicotinamide adenine dinucleotidephosphate (NADPH) by Photosystem I [9],

NADPþ þ Hþ þ 2e� ! NADPH: ð2ÞOur analogue of Photosystem I in heterogeneous catalysis is CO2

photoconversion using LDH and Pd/TiO2 to form methanol andmethane [6–8]:

CO2 þ 6Hþ þ 6e� ! CH3OHþ H2O ð3Þ

CO2 þ 8Hþ þ 8e� ! CH4 þ 2H2O ð4Þ

2H2Oþ 2NADPþ ! O2 þ 2NADPHþ 2Hþ ð1þ 2Þ

4H2Oþ 2CO2 ! 3O2 þ 2CH3OH ð1þ 3Þ

2H2Oþ CO2 ! 2O2 þ CH4 ð1þ 4ÞAlthough significant progress has been achieved by many research-ers in the photoconversion of CO2 into fuels [1–3], most studieshave evaluated the photoconversion of dissolved CO2 in aqueoussolution, namely Eqs. (1) and (3) or (1) and (4). For example, thecritical role of oxygen vacancy (OV) sites in CO2/H2O was suggested[10–13], and enhanced photoconversion of CO2 was reported at ele-

40 S. Kawamura et al. / Journal of Catalysis 345 (2017) 39–52

vated reaction pressure of CO2 and moisture [10,11] or liquid water[11,14–19]; however, most of the mechanisms are unknown. Thus,the systematic dependence of CO2 photoconversion on reactionpressure and an independent understanding of the reduction sitesand mechanism for CO2 conversion (Eqs. (3) and (4)) are still openfor discussion.

The present study screened semiconductor photocatalysts forefficient photoreduction of CO2 into fuels using H2 as a reductantat 0.12–0.80 MPa in order to clarify the role of Eq. (3) and/or (4).The band gap (Eg) values of the semiconductors were variedbetween 2.6 and 5.6 eV by employing ZrO2, g-C3N4, BiOCl, TiO2,and [Zn3Ga(OH)8]2CO3�mH2O LDH while keeping the conductionband (CB) minimum more negative than the reduction potential(E�) for CO2 to methane, CO, or methanol (�0.32 to �0.11 V againstthe standard hydrogen electrode (SHE)) [3,8]. The reaction pres-sure dependence for Reaction (3) and/or (4) was carefullyinvestigated.

Sastre et al. reported CO2 conversion using H2 and Ni-based cat-alysts irradiated by a solar simulator, and the rates were surpris-ingly as high as those using a photocatalyst: 55 mmol h�1 gcat�1

[20]. The temperature and gas pressure reached 423 K and somehundreds of kPa due to the exothermic nature of Eq. (4). On thebasis of blank tests below 453 K in the absence of light, photocat-alytic activation of H2 to form Ni–H species followed by thermalhydrogenation of CO2 to formate on metallic Ni was suggested[2,20]. The possibility of a thermal assist for activation is also dis-cussed in this study.

The reason that a volcano-type dependence of CO2 photocon-version on reaction pressure is obtained using the most activePd/TiO2 photocatalyst was investigated by monitoring the massbalance of reactions, the oxidation state of Pd, and the concentra-tion of OV sites in TiO2 by X-ray absorption fine structure (XAFS).This paper is the first of several in which the sites and mechanismfor reaction in CO2 + H2 are compared with those for reaction inCO2 + H2O. As site separation between photo-oxidation (H2 ormoisture) and photoreduction (CO2), and the concentrations androles of OV sites on the catalysis are complicated issues, the photo-conversion of CO2 using moisture is reported separately [21].

2. Experimental

2.1. Catalyst synthesis

An aqueous solution (1.0 mM) of Ag nitrate (>99.8%, Wako PureChemical) was stirred with ZrO2 {JRC-ZRO-3, Catalysis Society ofJapan, specific surface area (SA) 94 m2 g�1} for 24 h. An aqueoussolution of 40 mM of NaBH4 (>95%, Wako Pure Chemical) was thenadded in the molar ratio Ag:NaBH4 = 1:4 and the solution was stir-red for 1 h. The solution was filtered using a polytetrafluoroethene-based membrane filter (Omnipore JVWP04700, Millipore; pore size0.1 lm) and the collected yellow precipitate was washed withdeionized water (<0.055 lS cm�1) supplied by a model RFU424TA(Advantec). The powder was dried under vacuum at 290 K for24 h. The obtained yellow powder is denoted as Ag/ZrO2. The load-ing of Ag was 0.5 wt.% (Table 1a).

Ten grams of urea (>99%, Wako Pure Chemical) was heated at823 K for 3 h and the resultant yellow powder was washed with0.1 M of nitric acid (Wako Pure Chemical) and deionized water toobtain g-C3N4 [22].

BiOCl was synthesized via a solvothermal procedure describedin Ref. [23]. Briefly, 2.0 g of Bi nitrate pentahydrate (>99.5%, WakoPure Chemical) and 1.3 g of cetyltrimethylammonium chloride(>95%, Wako Pure Chemical) were dissolved in 80 mL of ethyleneglycol (EG, >99.5%, Wako Pure Chemical). The solution was stirredfor 1 h and 1.0 M of KOH (85%, Wako Pure Chemical) in EG was

added. The reaction solution was maintained at 433 K for 12 h.The solution was filtered and the collected white powder waswashed with deionized water and ethanol.

Sodium tetrachloropalladate (>98%, Sigma Aldrich) solution(1.0 mM) was stirred with TiO2 (P25, Degussa; anatase phase:rutilephase = 8:2, specific SA 60 m2 g�1) at 290 K for 24 h. Then 40 mMof NaBH4 aqueous solution was added in the molar ratio Pd:NaBH4 = 1:8 and the suspension was stirred at 290 K for 1 h. Thesolution was filtered using a JVWP04700 filter and the collectedyellow precipitate was washed with deionized water before dryingunder vacuum at 290 K for 24 h. The obtained gray powder wasdenoted as Pd/TiO2. The loading of Pd was 0.5 wt.% (Table 1d).

Cu(NO3)2�3H2O (0.078 g; >99.9%, Wako Pure Chemical) and urea(2.5 g) were dissolved in 50 mL of deionized water. A quantity of1.0 g of TiO2 (P25) was added to this solution. Thereafter, the tem-perature of the suspension was increased to 353 K and kept con-stant for 4 h while stirring as a deposition–precipitation step. Thesample was then washed with deionized water (50 mL each) andfiltered five times using a JVWP04700 filter. After drying, the sam-ple temperature was elevated at a rate of 1.25 K min�1 and kept at673 K in air for 2 h [24]. The loading of Cu was 3.2 wt.% (Table 1e).

The synthetic procedure for preparing LDH [Zn3Ga(OH)8]2CO3-�mH2O has been described previously [8]. A quantity of 1.0 mMof Ag nitrate solution was stirred with [Zn3Ga(OH)8]2CO3�mH2Opowder at 290 K for 24 h and then 40 mM NaBH4 aqueous solutionwas added in the molar ratio Ag:NaBH4 = 1:4 and stirred for 1 h.The solution was filtered using a JGWP04700 filter and the col-lected yellow precipitate was washed with deionized water. Theobtained yellow powder is denoted Ag/Zn3Ga-LDH. The loadingof Ag was 0.5 wt.% (Table 1f).

2.2. Characterization

X-ray diffraction (XRD) patterns were observed using a D8ADVANCE diffractometer (Bruker) at the Center for AnalyticalInstrumentation, Chiba University, at a Bragg angle (hB) of2hB = 5.0–60� with a scan step of 0.02� and a scan rate of 3 s perstep. The measurements were performed at 40 kV and 40 mA usingCu Ka emission (wavelength k = 0.15419 nm) [25,26] and a nickelfilter. Crystallite sizes (t) were estimated using the Scherrerequation

t ¼ 0:9kPeak width� coshB

: ð5Þ

Ultraviolet (UV)–visible spectra were measured using a Model V-650 (JASCO) spectrophotometer equipped with an integratingsphere (ISV-469, JASCO) for diffuse reflectance measurements inthe wavelength range 200–800 nm. The Eg values were estimatedvia simple extrapolation of the absorption edge or by fitting tothe Davis–Mott equation,

ahm / ðhm� EgÞn; ð6Þwhere a, h, and m are the absorption coefficient, Planck’s constant,and the frequency of light, respectively, and n is 1/2, 3/2, 2, or 3for allowed direct, forbidden direct, allowed indirect, and forbiddenindirect electronic transitions, respectively [8,27,28]. Fits to Eq. (6)were performed assuming each n value, or using the n value evalu-ated based on fits to the log–log form of Eq. (6) using the Eg valueobtained by simple extrapolation of the absorption edge to thewavelength axis in the UV–visible spectra.

Transmission electron microscopy (TEM) images were observedusing a Model H-7650 transmission electron microscope (Hitachi)at an accelerating voltage of 100 kV. A tungsten filament was usedin the electron gun and samples were mounted on conducting car-bon with a Cu grid mesh (150 mesh per inch). The magnification

Table 1Basic physicochemical properties of the photocatalysts screened in this study.

Entry Photocatalyst Metal loading(wt.%)

Color Peak dueto SPR (nm)

Pd mean particle size(nm)

Eg (eV) n obtained fromlog–log formof Eq. (6)

TEM EXAFS Extrapolated Based on Eq. (6) (associated n)

a Ag/ZrO2 0.5 Yellow 428 5.1 5.2 (n = 1/2) 1.255.0 (n = 3/2)

b g-C3N4 – Lime yellow 2.9 3.0 (n = 1/2) 0.992.8 (n = 3/2)

c BiOCl – White 3.6 3.8 (n = 1/2) 1.063.4 (n = 3/2)

d Pd/TiO2 0.5 Gray (Too broad) 3.1(±0.9) 1.1(±0.1) 3.1 3.4 (n = 1/2) 1.05{3.9(±1.3)a}

{2.3(±0.1)b}

2.9 (n = 3/2)

e Cu/TiO2 3.2 Light gray �440 3.2 3.4 (n = 1/2) 0.983.0 (n = 3/2)

f Ag/Zn3Ga-LDH

0.5 Yellow 395 5.2 5.6 (n = 1/2) 1.325.0 (n = 3/2)

a After photocatalytic tests under CO2 (0.09 MPa) and H2 (0.21 MPa) for 5 h.b After photocatalytic tests under CO2 (2.3 kPa) and H2 (22 kPa) for 5 h.

S. Kawamura et al. / Journal of Catalysis 345 (2017) 39–52 41

was between 60,000 and 200,000 times. Cross-sectional scanningelectron microscopy (SEM) images were also observed using aModel JSM-6510 scanning electron microscope (JEOL) at an accel-erating voltage of 20 kV. A tungsten filament was used in the elec-tron gun. The photocatalyst film on a Pyrex glass plate was cut andmounted on the Al sample holder by an adhesive. The incidentangle of electrons with reference to the normal of the sample sur-face was between 75� and 85�. The magnification was between 200and 3000 times.

Palladium K-edge XAFS spectra were measured at 290 K intransmission mode in the Photon Factory Advanced Ring at theHigh Energy Accelerator Research Organization (KEK, Tsukuba)on beamline NW10A [23,28,29]. The storage ring energy was6.5 GeV and the ring current was 52.1–20.7 mA. A Si (311)double-crystal monochromator and a Pt-coated cylindrical focus-ing mirror were inserted into the X-ray beam path. The X-rayintensity was maintained at 67% of the maximum flux using apiezo translator applied to the crystal to suppress higher harmon-ics. The slit opening was 1.0 mm (vertical) � 2.0 mm (horizontal)in front of the I0 ionization chamber. Samples (10 mg of film)mounted on Pyrex glass plates for photoconversion tests in0.12 MPa CO2 and 0.28 MPa H2 were measured at 290 K in the flu-orescence mode using a Lytle detector and a 3 lm thick RuO2 filterbetween the sample and the If ion chamber [30–32]. The Pd K-edgeabsorption energy was calibrated to 24,348 eV using the spectrumof 12.5 lm thick Pd foil [25].

Titanium K-edge XAFS spectra were measured in transmissionmode in the Photon Factory at KEK on beamline 9A [33,34]. Thestorage ring energy was 2.5 GeV and the ring current was 447.7–342.7 mA. A Si (111) double-crystal monochromator and a pairof bent conical mirrors were inserted into the X-ray beam path.Spectra for the Pd/TiO2 sample (10 mg) diluted by boron nitridewere measured under argon. The Ti K-edge absorption energywas calibrated to 4964.5 eV using the spectrum of 5 lm thick Tifoil [25].

XAFS data were analyzed using the XDAP software package[35]. The pre-edge background was approximated by a modifiedVictoreen function C2/E2 + C1/E + C0. The background of the post-edge oscillation was approximated by a smoothing spline functionand was calculated for the particular number of data points, using

XData Points

i¼1

ðlxi � BGiÞ2expð�0:075k2i Þ

6 smoothing factor; ð7Þ

where k is the angular wavenumber of the photoelectrons.

Multiple-shell curve-fit analyses were performed [23,28,29,33]for the Fourier-filtered k3-weighted extended XAFS (EXAFS) datain k- and R-space using empirical amplitudes extracted from theEXAFS data for Pd foil, PdO powder, and rutile-type TiO2 powder.The interatomic distance (R) and its associated coordination num-ber (N) for the Pd–Pd, Pd–O, Ti–O, and Ti(–O–)Ti pairs were set to0.27509 nm with an N value of 12 [36], 0.2026 nm with an N valueof 4 [37], 0.1959 nm with an N value of 6, and 0.3058 nm with an Nvalue of 12 [33,38,39], respectively. The many-body reduction fac-tor S02 was assumed to be equal for the sample and the reference.

The number of independent data points in the fit range was cal-culated based on the Nyquist theorem and the EXAFS-specificmodification by Stern [35],

m ¼ 2DkDRp

þ 2: ð8Þ

The number was 10.8–13.1 for Pd K-edge EXAFS data and 14.4–15.8for Ti K-edge EXAFS data. Three-shell fits were also tried for TiEXAFS; however, the fit errors were significantly greater than thoseobtained in two-shell fits. Thus, two-shell fits (four variables in ashell) were chosen for both Pd and Ti K-edge EXAFS analyses.

2.3. High-pressure photoconversion tests of CO2

The synthesized/prepared catalysts (Table 1a–f) were immersedin deionized water and agitated by ultrasound (430W, 38 kHz) for3 min. The suspension was mounted on a Pyrex glass plate(25 � 25 � 1 mm) and dried at 373 K for 12 h. The area of theobtained films was 20 � 20 mm. Films on plates or as-synthesized fine powder samples in a Pyrex dish (Uinternal = 37mm) were introduced into a homemade high-pressure stainlesssteel reactor equipped with quartz double windows, a pressuregauge, and Swagelok valves (Fig. 1A, B) for photoconversion tests.The effective internal volume of the reactor was 98.4 mL.

A mixed CO2/H2 gas (0.20–0.80 MPa in a ratio of 2:8, 3:7, 4:6, or5:5) was introduced into the reactor. The photocatalyst was irradi-ated with UV–visible light provided by a 500-W Xe arc lamp(Model OPM2-502, Ushio) through quartz windows in the reactorchamber for 5 h (Fig. 1A, B). The distance between the light exitand the photocatalyst surface was 82.7 mm. Light transmissionwas checked using a photosensor and a counter (Model PCM-01,Prede and Model KADEC-UP, North One, respectively). The lightintensity at the center of the photocatalyst was 90.2 mW cm�2

(Fig. 2a). The quartz windows of the high-pressure reactorabsorbed/reflected 9.5% of the light (Fig. 2b), and 100 mg of Pd/

Fig. 1. (A, B) High-pressure stainless reactor equipped with quartz windows and apressure gauge for CO2 photoconversion tests irradiated by UV–visible light from a500-W Xe arc lamp (A: top view, B: side view). (C) Photocatalyst films(20 � 20 mm) of Ag/ZrO2, g-C3N4, BiOCl, Pd/TiO2, Cu/TiO2, and Ag/Zn3Ga-LDH (topto bottom).

42 S. Kawamura et al. / Journal of Catalysis 345 (2017) 39–52

TiO2 powder on a Pyrex dish and 10 mg of Pd/TiO2 film on Pyrexglass plate absorbed/reflected/scattered 40% and 49% of the light,respectively (Fig. 2c, d). Densely mounted films were advantageousfor light absorption efficiency compared with fine powders basedon the difference of their mean areal densities of 2.5 versus9.3 mg cm�2, respectively, and the exposed ratio of particles. Finepowders should reflect/scatter more at exposed faces and less lightreach at the lower part of layers (Fig. 2c) compared with denselymounted films (Fig. 2d).

As comparisons, photocatalytic tests were also performed usingan L42 filter (2.5 mm thick, Kenko) that filters the light less thank = 420 nm [40] or a U-330 filter (2.5 mm thick, Kenko) that filtersthe light less than k = 250 nm and between k = 390 nm and 695 nm[41], set between the light exit and the high-pressure reactor(Fig. 1A, B).

Fig. 2. The light intensity at the photocatalyst position in the high-pressure stainless-stepresence of the quartz windows and 100 mg of Pd/TiO2 powder (c), and in the presence

After 5 h of irradiation, the reaction gas was analyzed usingpacked columns of 13X-S molecular sieves and polyethylene glycol(PEG-6000) supported on Flusin P (GL Sciences) set in a gas chro-matograph equipped with a thermal conductivity detector (ModelGC-8A, Shimadzu) [6–8,28,42]. Helium (>99.99995%) was used as acarrier gas for all analytes except H2, for which argon (>99.998%)was used. The amounts of methanol and moisture were double-checked by concentrating them using a trap made from a mixtureof diethyl ether and dry ice (192 K) to separate them from H2 andmost of the CO2.

The experimental errors of 4 Pa of methanol or moisture wereevaluated in this GC analysis procedure. The errors were 7.7%and 8.4%, but increased to 41% and 26% by the adsorption ofmethanol or moisture onto the surface of the Pyrex glass tubeand vacuum grease (Apiezon H). The gas adsorption could be min-imized by completely heating the vacuum system using a dryerjust before the online GC analysis.

Blank tests were also performed by eliminating one of the threecontrol factors or adding one alternative control factor: (i) UV–vis-ible light, (ii) reactants, (iii) catalyst, and (iv) temperature, i.e., in theabsence of any light at 298 or 353 K, in 0.12 MPa of CO2 only, in0.28 MPa of H2 only, or 0.4 MPa of helium (the absence of any reac-tants) and in the absence of any catalyst.

3. Results

3.1. Characterization by XRD and UV–visible spectroscopy

The syntheses of g-C3N4, BiOCl, and Zn3Ga-LDH and the prepa-ration of Ag/ZrO2, Pd/TiO2, Cu/TiO2, and Ag/Zn3Ga-LDH were veri-fied by their XRD patterns (Fig. 3) and UV–visible spectra (Fig. 4).For Ag/ZrO2 (Fig. 3a), the XRD peaks appeared at 2hB = 17.5�,24.3�, 28.3�, 31.5�, 34.3�, 35.3�, 38.6�, 41.0�, 45.1�, 50.3�, 54.2�,and 55.6� and were ascribed to 001, 011, 1 1 1, 111, 020, 002,120, 1 1 2, 2 0 2, 022, 003, and 310 reflections of monoclinicZrO2 [43,44]. No peaks due to Ag metal or Ag2O nanoparticles wereobserved [28,29].

Broad peaks at 2hB = 13.3� and 27.5� and a weak peak at2hB = 44� in pattern b appeared essentially at the same positionas those reported for g-C3N4 in the literature [22,45] and wereassigned to 100, 002, and 200 reflections, respectively. The peakpositions and relative intensities for peaks at 12.0�, 26.0�, 32.7�,33.6�, 41.0�, 46.8�, 49.8�, 54.3�, and 58.8� in pattern c were in

el reactor (Fig. 1) (a), in the presence of the quartz windows of the reactor (b), in theof the quartz windows and 10 mg of Pd/TiO2 film (d).

Fig. 3. XRD patterns for Ag/ZrO2 (a), g-C3N4 (b), BiOCl (c), Pd/TiO2 (d), and Ag/Zn3Ga-LDH (e).

Fig. 4. UV–visible spectra for fresh samples of Ag/ZrO2 (a), g-C3N4 (b), BiOCl (c), Pd/TiO2 (d), Cu/TiO2 (e), and Ag/Zn3Ga-LDH (f).

S. Kawamura et al. / Journal of Catalysis 345 (2017) 39–52 43

accord with the literature reports for BiOCl and were assigned to001, 101, 110, 102, 112, 200, 113, 202, and 212 reflections,respectively, of the tetragonal layered structure [23].

In the pattern for Pd/TiO2 (Fig. 3d), peaks appeared at2hB = 25.4�, 37.0�, 38.6�, 48.1�, 54.0�, and 55.1� and were assignedto the 101, 103, 112, 200, 105, and 211 reflections, respectively,of the anatase phase, and the peaks that appeared at 2hB = 27.5�,36.1�, 37.9�, 41.3�, 44.1�, and 56.7� were assigned to the 110,101, 004, 111, 210, and 220 reflections, respectively, of the rutilephase [46,47]. No peaks arising from Pd nanoparticles wereobserved above the detection limit.

As for pattern e for Ag/Zn3Ga-LDH, peaks appeared at2hB = 11.7�, 23.5�, 33.5�, 34.2�, 36.9�, 38.9�, 43.6�, 46.4�, and 52.6�and were assigned to the 003, 006, 009, 012, 104, 015, 017,018, and 100 reflections of Zn3Ga-LDH, respectively, in accordwith the literature [7,8].

For Ag/ZrO2, the UV absorption edge was extrapolated to242 nm, corresponding to an Eg value of 5.1 eV (Table 1a). A broad

peak centered at 428 nm appeared due to the surface plasmon res-onance (SPR) of the Ag nanoparticles (Fig. 4a). An SPR peak at411 nm was reported for Ag particles with a mean size of 1.6 nmon LDH [28].

The absorption edge shifted from the UV to the visible lightregion (435 nm) for g-C3N4 (Fig. 4b), corresponding to an Eg valueof 2.9 eV (Table 1b). The absorption edge in the UV region wasextrapolated to 344 nm, corresponding to an Eg value of 3.6 eVfor BiOCl, in agreement with the literature value (Table 1c andFig. 4c) [23].

The absorption edges were extrapolated to 394 and 384 nm,corresponding to Eg values of 3.1 and 3.2 eV for Pd/TiO2 and Cu/TiO2 (spectra d and e), respectively, in agreement with the litera-ture for TiO2 [33]. Nearly flat background absorption extended overthe entire range of visible light due to various sizes of Pd nanopar-ticles, and a weak peak appeared at approximately 440 nm due toSPR effects arising from the Cu.

3.2. TEM and cross-sectional SEM observations

TEM images were observed for the Pd/TiO2 photocatalyst. Forthe fresh sample, polygonal particles of TiO2 were observed withsizes distributed between 30 and 50 nm (Fig. 5A). Smaller particlesthat are clearly darker than the TiO2 were identified as Pd speciesand their mean size was 3.1 ± 0.9 nm (Fig. 5A, histogram andTable 1d). The wavelength of the SPR peaks progressivelydecreased as the metal particle size decreased due to the changein the oscillation frequency of confined free electrons [48]. The sizedistribution of the Pd species was between 1.0 and 6.0 nm, leadingto SPR peaks at various wavelengths with broad absorbance in thewavelength range 400–800 nm (Fig. 4d).

TEM images were also observed for Pd/TiO2 samples after 5 h ofCO2 photoconversion testing in CO2 (0.09 MPa) and H2 (0.21 MPa).The particle size distribution of Pd species slightly widened to 1–9 nm and the mean size increased by 0.8 nm to 3.9 ± 1.3 nm(Fig. 5B, histogram and Table 1d).

SEM images revealed that the surface of fresh Pd/TiO2 photocat-alyst films coated on Pyrex glass was quite flat and smooth(Fig. 5C1). Cross-sectional views were also observed (Fig. 5C2)and the mean thickness of the photocatalyst films was 11 lm.The mean particle size of the TiO2 was estimated to be 32 nm,based on analysis of the 101 XRD peak of the anatase phase usingEq. (5), which is in agreement with the size range 30–50 nmobserved by TEM (Fig. 5A).

3.3. Monitoring the active sites using Pd and Ti K-edge XAFS

First, Pd K-edge XAFS spectra were measured for fresh Pd/TiO2

(100 mg) and Pd/TiO2 after testing in CO2 (2.3 kPa) + H2 (22 kPa)under UV–visible light irradiation for 5 h. The photocatalytic testconditions were conducted at lower than atmospheric pressure[6–8,28]. The X-ray absorption near-edge structure (XANES) spec-tra are depicted in Fig. 6A–c and d, respectively. The energy valuesat the tops and the bottoms of the post-edge oscillation for normal-ized oscillation lt coincided with those for Pd metal (Fig. 6A-b);however, the pattern was quite different from that of PdO(Fig. 6A-a), demonstrating the dominant metallic Pd0 state for freshPd/TiO2 samples and for those after CO2 photoconversion tests at24 kPa. The amplitude of the post-edge oscillation was apparentlylower for Pd/TiO2 (Fig. 6A-c, d) than that for Pd foil (spectrum b),indicating a Pd particle size of a few nanometers, as already shownby TEM (Fig. 5A, B).

The best-fit results and fit errors of the Pd K-edge EXAFS for Pd/TiO2 photocatalysts are summarized in Table 2. The data were wellfitted with two shells: Pd–O and Pd–Pd for fresh Pd/TiO2 (Fig. 6B-c3, c4, and Table 2-c). The coordination of the Pd–Pd interatomic

Fig. 5. TEM images of Pd/TiO2 as prepared (A) and after CO2 photoreduction tests in CO2 (0.09 MPa) and H2 (0.21 MPa) (B), and cross-sectional SEM images of fresh Pd/TiO2

(C1, C2).

44 S. Kawamura et al. / Journal of Catalysis 345 (2017) 39–52

pair at 0.276 nm (Fig. 6B-c2) was predominant (N = 5.9), which isconsistent with the assignment of the predominant Pd0 based onthe XANES data above (Fig. 6A–c). The minor coordination of Pd–O (N = 1.9) arises from the interfacial bonding between metallicPd nanoparticles and the TiO2 surface [49–53]. The N value forPd–Pd (5.9) corresponds to a mean nanoparticle size of1.1 ± 0.1 nm (Table 1d) [53]. The discrepancy compared with themean particle size based on TEM (3.1 ± 0.9 nm) is found becauseparticles smaller than �1 nm cannot be observed by TEM.

For the samples measured after undergoing photoconversionunder 24 kPa and light irradiation (see Photoconversion in CO2 andH2 at 0.20–0.80 MPa), the major Pd–Pd interatomic distance wasalso 0.276 nm (Table 2-d and Fig. 6B-d2, d4). The N value slightlyincreased by 3.0 relative to the fresh sample (Table 2-c, d). Thisvalue corresponds to the mean particle size of 2.3 ± 0.1 nm, whichis 1.2 nm larger than for the fresh sample [53]. The particle growthbased on EXAFS was consistent with the corresponding value(0.8 nm) derived from TEM (Table 1d). The Pd–O interatomic dis-tance increased by 0.01–0.21 nm and the N value decreased signif-icantly from 1.9 to 0.57. Presumably, the reason for this is that thenumber of OV sites increased at the interface between Pd and theTiO2 surface and/or the relative number of interface Pd sites in con-tact with the TiO2 surface decreased due to partial growth of the Pdparticles by 1 nm [49–53].

After 0.40 MPa of CO2 photoconversion testing using 100 mg ofPd/TiO2, the peak energy and intensity in XANES were quite similar

to those of fresh samples (Fig. 6A–c, e). The N value of the Pd–Pdpair was 5.0 and did not significantly change from 5.9 for the freshsample based on the fit errors (Table 2-c, e and Fig. 6B-e3, e4),whereas the N value for the Pd–O pair significantly decreased from1.9 to 0.9 (Table 2-c, e). Thus, the increase in OV sites for a nearlyconstant mean size of Pd particles during the photoconversion testwas suggested after the photoconversion test at 0.40 MPa. The rea-son that the mean Pd particle size increased only under lower pres-sure conditions but remained unchanged under higher pressureconditions (Table 2-d, e) is unknown; however, the strong adsorp-tion of H on Pd [54] may have prevented the coalescence of Pdnanoparticles. The XANES spectrum (Fig. 6A–e) resembled that offresh Pd/TiO2 (c), supporting the interpretation based on EXAFS.

The XANES spectrum taken for 10 mg of Pd/TiO2 film (Fig. 1C)after a photoconversion test at 0.40 MPa (Fig. 6A–e0) was quite sim-ilar to the spectrum in Fig. 6A–c for a fresh sample. In summary, Pdsites in 10–100 mg of Pd/TiO2 photocatalyst remained Pd0 sitesbefore and after photoreduction tests in CO2 and H2 over a pressurerange from 24 kPa to 0.40 MPa.

To compliment Pd K-edge XAFS, Ti K-edge EXAFS was measuredfor 10 mg of Pd/TiO2 photocatalyst after the photoconversion testat 0.40 MPa. The N(Ti–O) and N(Ti(–O–)Ti) values were 6.1 and11 (Table 3-e0 and Fig. 6C–e03, e04), similar to the values of 6 and12 for the untreated rutile or anatase phase TiO2, respectively[33,36,37]. In contrast, the decrease in OV sites at the interfacebetween Pd and TiO2 surface sites under these conditions was sig-

Fig. 6. Pd K-edge XANES (A) and EXAFS (B) and Ti K-edge EXAFS (C) for PdO (a), Pd foil (thickness 12.5 lm; b), fresh Pd/TiO2 (c), and Pd/TiO2 after 5 h photocatalytic testing inCO2 (2.3 kPa) and H2 (22 kPa) using 100 mg of Pd/TiO2 (d) and in CO2 (0.12 MPa) and H2 (0.28 MPa) using 100 mg (e) or 10 mg of Pd/TiO2 (e0). Spectra were measured in thetransmission mode except that the Pd K-edge of sample e0 was taken in the fluorescence mode. Panels in B and C: k3-weighted EXAFS oscillation (1), its associated Fouriertransform (2), and best-fit results in k-space (3) and R-space (4).

S. Kawamura et al. / Journal of Catalysis 345 (2017) 39–52 45

Table 2Best-fit results of Pd K-edge and EXAFS spectra for Pd/TiO2 photocatalysts.

Entry Samples Pd–O Pd–Pd Goodness of fitR (nm) R (nm)N ND(r2) (10�5 nm2) D(r2) (10�5 nm2)

c Fresh 0.199 (±0.006) 0.276 (±0.001) 1.2 � 103

1.9 (±0.4) 5.9 (±0.5)�4.8 (±1.7) 5.0 (±0.6)

da 5-h photocatalytic test under CO2 (2.3 kPa) + H2 (22 kPa), 100 mg 0.21 (±0.04) 0.276 (±0.002) 2.0 � 102

0.57 (±0.2) 8.9 (±0.4)�7.9 (±1.7) 3.3 (±0.3)

eb 5-h photocatalytic test under CO2 (0.12 MPa) + H2 (0.28 MPa), 100 mg 0.2018 (±0.0004) 0.2748 (±0.0003) 1.1 � 103

0.9 (±0.3) 5.0 (±0.4)�6 (±2) 2.6 (±0.3)

a After photocatalytic test listed in Table 4a.b After photocatalytic test listed in Table 4f.

Table 3Best-fit results of Ti K-edge EXAFS spectra for Pd/TiO2 photocatalysts.

Entry Samples Ti–O Ti(–O–)Ti Goodness of fitR (nm) R (nm)N ND(r2) D(r2)(10�5 nm2) (10�5 nm2)

e0a 5-h photocatalytic test under CO2 (0.12 MPa) + H2 (0.28 MPa), 10 mg 0.1977 (±0.0002) 0.313 (±0.002) 9.9 � 104

6.1 (±0.4) 11 (±2)1.6 (±0.4) �2 (±1)

a After photocatalytic test listed in Table 4n.

46 S. Kawamura et al. / Journal of Catalysis 345 (2017) 39–52

nificant based on the N(Pd–O) decrease (Table 2-c, e) and should bepopulated only at the interface with Pd nanoparticles after thephotoconversion test. The Ov sites were negligible in/on TiO2 apartfrom Pd nanoparticles.

3.4. Photoconversion in CO2 and H2 at 0.20–0.80 MPa

CO2 photoreduction using H2 and 100 mg of Pd/TiO2 or Ag/Zn3-Ga-LDH powder at a reaction pressure of 24 kPa [6–8] irradiatedwith UV–visible light produced methane, CO, and/or methanol,and the rates on a C basis were 0.14–1.0 lmol h�1 gcat�1

(Table 4a, b). Table 4 lists normalized formation rates per photocat-alyst weight for the practical comparisons to some of the literature,e.g., Refs. [10–12,14–17,55,56], but real formation rates in the unitlmol h�1 are also listed in Table S1 in the Supplementary Informa-tion for the comparisons with the other part of the literaturebecause photocatalyst weight had a critical influence on the photo-catalytic activities in this section. If the reaction pressures wereelevated to 0.12 MPa of CO2 and 0.28 MPa of H2, the total ratesincreased by a factor of 1.8–2.6 times (0.36–1.8 lmol h�1 gcat�1;Table 4f, h). At a total reaction pressure of 0.40 MPa, the order ofthe total photoformation rates using 100 mg of powder sampleson a C basis was as follows (Tables 4 and S1):

Cu=TiO2ð2:8Þ > Pd=TiO2ð1:8Þ >> Ag=ZrO2ð0:64Þ� g� C3N4ð0:53Þ > Ag=Zn3Ga� LDHð0:36Þ> BiOClð0:25Þ: ð9Þ

Ag/ZrO2, g-C3N4, and BiOCl mostly produced CO at formation ratesof 0.64, 0.53, and 0.15 lmol h�1 gcat�1, respectively (Table 4c–e),whereas Pd/TiO2, Cu/TiO2, and Ag/Zn3Ga-LDH favored methane for-mation at rates of 1.7, 2.8, and 0.32 lmol h�1 gcat�1, respectively(Table 4f–h). A minor product was methanol, which was formedat a rate of up to 0.003 lmol h�1 gcat�1 using these photocatalysts.

The theoretical formation rate of water, r(H2O), is given by theequation

rðH2OÞ ¼ 2rðCH4Þ þ rðCOÞ þ rðCH3OHÞ ð10Þbased on the stoichiometry of the equations

CO2 þ 4H2 ! CH4 þ 2H2O ð11Þ

CO2 þ H2 ! COþ H2O ð12Þ

CO2 þ 3H2 ! CH3OHþ H2O ð13Þwhich account for the formation of methane, CO, and methanol.

For Ag/ZrO2, g-C3N4, Cu/TiO2, and Ag/Zn3Ga-LDH, the r(H2O)values were calculated to be 0.64, 0.53, 5.6, and 0.68 lmol-H2O h�1 gcat�1, which are in good agreement with the experimentalvalues of 0.64, 0.57, 6.3, and 0.65 lmol-H2O h�1 gcat�1, respectively(Table 4c, d, g, h) if compared with the maximum evaluation errorof H2O by GC (26%).

Conversely, the observed water formation rate using BiOCl(1.3 lmol h�1 gcat�1, Table 4e) was larger by a factor of 3.7 than ther(H2O) value based on Eq. (10) (0.35 lmol h�1 gcat�1), suggesting con-current noncatalytic hydrogenation of reactive lattice O atoms atthe BiOCl surface [23,57]. Using Pd/TiO2, the observed water for-mation rate was 2.2 lmol h�1 gcat�1 (Table 4f), which is a factor of0.63 times lower than the r(H2O) value (3.5 lmol h�1 gcat�1) thatwas a significant difference in comparison to maximum evaluationerror by GC (26%). Because Pd nanoparticles remained metallic andthe number of OV sites at the interface with the TiO2 surfaceincreased under the photocatalytic conditions (Table 2-e andFig. 6B–e), approximately 40% of O originating from CO2 shouldbe adsorbed onto the TiO2 surface away from Pd nanoparticles.The quantity of missing O atoms based on the stoichiometry ofEqs. (11) and (12) was 0.65 lmol. The quantity of hydroxy groupsin the untreated Pd/TiO2 catalyst (7 species nm�2) [58] amountedto 69 lmol in 100 mg of Pd/TiO2 powder. These hydroxy groupsare plausible adsorption sites under the photocatalytic conditionsif 0.94% of them desorb as water, leaving behind OV sites duringthe pretreatment of Pd/TiO2.

Table 4Dependence of CO2 photoconversion on the amount of photocatalyst and reactant pressure, and control conversion results in the absence of UV–visible light, catalyst, or reactant.

Ent. Photocatalyst Reactants (MPa) Formation rates (lmol h�1 gcat�1)

Type Wt (mg) CO2 H2 CO CH3OH CH4 H2O O2P

Ca

a Pd/TiO2 100 0.0023 0.022 <0.008 <0.0004 1.0 5.0 0 1.0b Ag/Zn3Ga-LDH 0.11 0.028 <0.012 – 0.14c Ag/ZrO2 0.12 0.28 0.64 0.0021 <0.012 0.64 0.64d g-C3N4 0.53 0.0021 <0.012 0.57 0.53e BiOCl 0.15 <0.0004 0.10 1.3 0.25f Pd/TiO2 0.057 <0.0004 1.7 2.2 1.8g Cu/TiO2 <0.008 0.0030 2.8 6.3 2.8h Ag/Zn3Ga-LDH 0.044 <0.0004 0.32 0.65 0.36i Ag/ZrO2 10 <0.08 <0.004 4.3 6.5 4.3j g-C3N4 <0.08 <0.004 3.8 4.1 3.8k BiOCl 1.9 <0.004 1.1 37 3.0l Pd/TiO2 0.06 0.14 <0.08 <0.004 13 38 13m 0.09 0.21 <0.08 0.18 23 43 23n 0.12 0.28 <0.08 <0.004 38 68 38n0b <0.08 0.018 8.8 11 8.8n00c <0.08 0.009 <0.12 7.2 <0.21n0 00d <0.08 <0.004 3.0 5.1 3.1n00 00e <0.08 0.016 1.3 9.8 1.3n00 00 0 0 0.28 <0.08 <0.004 <0.12 11 <0.20n00 00 00 0.12 0 <0.08 0.008 <0.12 19 <0.20n00 00 00 0 f 0 0 <0.08 <0.004 <0.12 8.2 <0.20o 0.15 0.35 <0.08 0.17 20 28 20p 0.18 0.42 <0.08 <0.004 14 12 14q 0.24 0.56 <0.08 <0.004 8.6 9.6 8.6r 0.12 0.12 <0.08 0.013 12 25 12s 0.28 0.28 <0.08 0.020 22 58 22t 0.07 0.28 <0.08 0.14 27 44 27u 0.19 0.28 <0.08 0.08 26 39 26v Cu/TiO2 0.12 0.28 <0.08 0.056 19 33 19w TiO2 <0.08 0.027 4.0 56 4.0x Ag/Zn3Ga-LDH 0.95 0.10 2.1 5.3 3.2y No catalyst 0 <0.0008g <0.00004g <0.0012g <0.03g <0.0020g

a Total molar amount of C-containing products.b Using U330 filter at the light exit (Fig. 1B).c Using L42 filter at the light exit (Fig. 1B).d In the absence of UV–visible light at 298 K.e In the absence of UV–visible light at 353 K.f In 0.40 MPa of helium.g In units of lmol h�1.

S. Kawamura et al. / Journal of Catalysis 345 (2017) 39–52 47

It seems contradictory that this effect of O uptake derived fromCO2 on OV sites on TiO2 was not confirmed using Cu/TiO2

(Table 4g). One of the reasons for the difference is the differentfinal treatment temperature: 290 K under vacuum (Pd/TiO2) versus673 K in air (Cu/TiO2) leading to fewer hydroxy groups for Cu/TiO2

[58].Next, at the same reaction pressure (0.40 MPa), the amount of

photocatalyst was decreased to 10 mg. Using 10 mg of Pd/TiO2 film,methane was exclusively formed at a rate of 38 lmol h�1 gcat�1,higher by a factor of 22 times than when 100 mg of Pd/TiO2 powderwas used (Table 4f, n). Samples of 10 mg of Ag/ZrO2, g-C3N4,Cu/TiO2, and Ag/Zn3Ga-LDH films exclusively (Table 4i, j, v) or pref-erentially (Table 4x) formed methane at rates of 4.3, 3.8, 19, and2.1 lmol h�1 gcat�1, factors of 6.7, 7.2, 6.8, and 8.9 times higher,respectively, on a total C basis than when 100 mg of the corre-sponding catalyst was used (Table 4c, d, g, h). A quantity of 10 mgof BiOCl film formed CO (64% selectivity) at a total formation rateof 3.0 lmol h�1 gcat�1 (Table 4k), higher by a factor of 12 than when100 mg of BiOCl was used (Table 4e).

The total photoformation rates on a C basis using 10 mg of pho-tocatalyst at 0.40 MPa of CO2 and H2 (Table 4j, k, n, v, x andTable S1) were in the order

Pd=TiO2ð38Þ > Cu=TiO2ð19Þ � Ag=ZrO2ð4:3Þ� g� C3N4ð3:8Þ > Ag=Zn3Ga� LDHð3:2Þ� BiOClð3:0Þ: ð14Þ

Formation rates were boosted by a factor of 22–6.7 times when10 mg of photocatalyst films was used; however, the activity orderdid not change compared with Eq. (9) using 100 mg of catalyst,except for a reversal of the order of the top two TiO2-basedcatalysts.

One of the reasons for these rate increases was the improve-ment of light absorption efficiency for the photocatalyst film com-pared with that for powders: 49% versus 40% (Fig. 2c, d) due to thedifference in the exposed ratio of particles. As noted in Section 2.3,the film of areal density of 2.5 mg cm�2 absorbed more light com-pared to the powders of 9.3 mg cm�2 due to the relatively greaterreflection/scattering by the powders. Furthermore, the activityincrease by a factor of 22–6.7 times using six kinds of photocata-lysts suggested that the other control factor(s) also exist.

Mass transfer limitation was suggested for catalyst pelletsgreater than 250 lm for the Fischer–Tropsch reaction [59]. A mean0.7 lm-thick TiO2 catalyst film was not controlled by mass transferlimitation for photocatalytic decomposition of formic acid [60]. Nomass transport effect was observed for 2–5 lm of Pt/Al2O3 catalystfilm for the CO preferential oxidation reaction [61]. Taking thelower reaction rates for photocatalytic CO2 conversion comparedwith these thermodynamically easier reactions [59–61], the masstransfer limitation seems negligible in the thin 10 mg films (thick-ness �11 lm, Fig. 5C2) and would not be significant in 100 mg ofpowder (�100 lm).

Instead, the anatase 1 0 1 and rutile 1 1 0 diffraction peaks,especially the anatase 1 0 1 peak, were relatively weak in the

48 S. Kawamura et al. / Journal of Catalysis 345 (2017) 39–52

XRD pattern for 10 mg of Pd/TiO2 film in comparison to ones forPd/TiO2 powder (Figs. S1 and 3d and Table S2). The population ofthese planes in parallel with the sample holder of the XRD appara-tus would be smaller and the other plane(s) favorable for CO2 pho-toreduction may be preferably exposed on Pd/TiO2 film (seeSupplementary Material).

Using 10 mg of Ag/ZrO2, g-C3N4, BiOCl, Pd/TiO2, Cu/TiO2, and Ag/Zn3Ga-LDH, r(H2O), formation rates were predicted to be 8.6, 7.6,4.1, 76, 38, and 5.3 lmol h�1 gcat�1 based on Eq. (10), and experimen-tal values were 6.5, 4.1, 37, 68, 33, and 5.3 lmol h�1 gcat�1, respec-tively (Table 4i–k, n, v, x). Thus, CO2 conversion using Ag/ZrO2,Pd/TiO2, Cu/TiO2, and Ag/Zn3Ga-LDH almost/exactly followed thestoichiometry of Eqs. (11-13) because the deviation between pre-dicted and experimental values was 11–24% smaller than the max-imum evaluation error for 4 Pa of water using GC (see Section 2).This result is consistent with the exclusive Pd0 sites of Pd/TiO2

(10 mg) before and after the photoreduction tests in CO2 and H2

at 0.40 MPa (Fig. 6A–e0).Using Pd/TiO2, the amount of methane formed at 0.40 MPa in

5 h was 1.9 lmol (Table 4n), whereas the number of surface Pdatoms in 10 mg of Pd/TiO2 catalyst was 0.19 lmol if we assumethe Pd dispersion is 0.40 for a mean nanoparticle size of 3.1 nm(Table 1d) [62], indicating a turnover of 10 and thus demonstratingcatalytic phenomena.

Using g-C3N4, the observed water formation rate was 0.54 timeslower than the calculated r(H2O) value; thus, approximately 50% ofO originating from CO2 would remain, or formed water readsorbedon the g-C3N4 surface. The water formation rate using BiOCl was afactor of 9.0 times higher than the r(H2O) value, suggesting concur-rent noncatalytic hydrogenation of reactive lattice O atoms at theBiOCl surface [23,57].

The possibility of reaction of CO2 with the product H2O was alsoconsidered. At the end of the photocatalytic test under 0.12 MPa ofCO2 + 0.28 MPa of H2 for 5 h using the most active Pd/TiO2, theamount of formed water was 68 lmol h�1 gcat�1 � 5 h �0.010 gcat = 3.4 lmol (Table 4n). In our preliminary photocatalytictest under 0.12 MPa of CO2 + 2.3 kPa (94 lmol) of moisture usingthe same catalyst (10 mg), the predominantmethane formation ratewas 14 lmol h�1 gcat�1 [21]. If the reaction for methane formation isassumed to be first-order in moisture, the rate under 0.12 MPa ofCO2 + 3.4 lmol of moisture would be 0.51 lmol h�1 gcat�1 at the finalstage of the photocatalytic test under CO2 + H2, which is negligiblecompared with the observed methane formation rate 38 lmol h�1

gcat�1 (Table 4n). Thus, the contribution of the reaction with theproduct H2O was negligible in the photocatalytic tests under high-pressure CO2 + H2.

Wavelength dependence of irradiation [6,28,63–65] for the Pd/TiO2 photocatalyst was also studied using filters between the lightsource and the high-pressure reactor (Fig. 1A, B). Using a filter ofU330 that cuts visible light and part of infrared light, the methaneformation rate was decreased to 23% of that under full irradiationfrom the Xe arc lamp (Table 4n0). Conversely, using a filter of L42that totally cuts UV light, methane was not detected above thedetection limit of GC but methanol was formed very slowly(0.009 lmol h�1 gcat�1; Table 4n00). Thus, the methane formation rateunder UV–visible light (Table 4n) was not the summation of thatunder UV light and that under visible light, but a synergetic effectof UV and visible light was suggested.

Furthermore, the dependence of reaction rates on total pressurewas studied using the most active Pd/TiO2 film in Eq. (14) at pres-sures in the range 0.20–0.80 MPa (Table 4l–q). The product wasexclusively methane (Fig. 7). The reaction order of methane forma-tion was 1.53, as determined by a log–log plot in the pressurerange 0–0.40 MPa. Methane formation rates reached a maximumof 38 lmol h�1 gcat�1 at 0.40 MPa and then suddenly dropped above

0.40 MPa and finally decreased to 8.6 lmol h�1 gcat�1 at 0.80 MPa(Table 4n, q and Fig. 7).

At total reactant pressures of 0.20, 0.30, 0.40, 0.50, 0.60, and0.80 MPa, r(H2O) values based on Eq. (10) were 26, 46, 76, 40, 28,and 17 lmol h�1 gcat�1, and the observed formation rates were 38,43, 68, 28, 12, and 9.6 lmol h�1 gcat�1, respectively (Table 4l–q).Thus, the photocatalytic tests nicely followed the stoichiometryat total reaction pressures of 0.30 and 0.40 MPa. Based on Pd K-edge and Ti K-edge XAFS results at 0.40 MPa of CO2 and H2 forPd/TiO2, Pd nanoparticles remained metallic and OV sites were onlyplausible at the interface between Pd and the TiO2 surface, butwere not widely distributed over the TiO2 surface (Fig. 6A–e, e0,B-e, C-e0 and Tables 2-e and 3-e0). Hence, at 0.20, 0.50, 0.60, and0.80 MPa, approximately 32%, 30%, 57%, and 44% of O, respectively,originating from CO2 would be adsorbed onto the TiO2 surfaceaway from Pd nanoparticles. The number of O atoms lacking basedon the stoichiometry of Eqs. (11) and (13) was 0.37–0.80 lmolafter the photoconversion tests at 0.50–0.80 MPa. The hydroxygroups of the untreated Pd/TiO2 catalyst (7 species nm�2) [58] cor-responded to 6.9 lmol in 10 mg of Pd/TiO2 film and are plausibleadsorption sites under the photocatalytic conditions if 5.4–12% ofhydroxy groups desorb as water to leave OV sites during the pre-treatment of Pd/TiO2. The adsorption of O originating from CO2

on the TiO2 surface would be more pronounced at relatively lowerphotocatalytic activity at 0.20, 0.50, 0.60, and 0.80 MPa.

Next, the pressure dependence was extended to create a two-dimensional contour plot of total C formation rates, which wereessentially equivalent to the methane formation rates, as a func-tion of H2 and CO2 partial pressures (Table 4l-s and Fig. 8). Themethane formation curve in Fig. 7 corresponds to a straight linefollowing PCO2 ¼ ð3=7ÞPH2 in Fig. 8. Photocatalytic tests were alsoperformed on three straight lines of PCO2 ¼ PH2 , PH2 = 0.28 MPa,and PCO2 = 0.12 MPa (Table 4r, s; Table 4n, n00 00, s, t, u; Table 4n, n00 00 0,r, respectively). The exclusive methane formation rates primarilydepended on PH2 ; the rates increased until PH2 reached 0.28 MPaand then gradually dropped until PH2 increased to 0.56 MPa,whereas they depended negligibly on PCO2 . However, within thezone PH2 = 0.20–0.35 MPa, the rates critically increased until aPCO2 of 0.12 MPa was reached and then gradually decreased untilPCO2 increased to 0.28 MPa (Fig. 8). In this addition of data underboth CO2 and H2 (Table 4r–u), predicted and observed amountsof water were in agreement within the maximum evaluation error(26%).

Blank tests were performed by controlling four factors (see Sec-tion 2) to compare with the photoconversion tests, as listed inTable 4n using 10 mg of Pd/TiO2. In the absence of reactants or inthe absence of catalyst (Table 4n00 00 00 0, y), no carbon-containingproducts were detected after 5 h of irradiation under UV–visiblelight. Water formation at a rate of 8.2 lmol h�1 gcat�1 in the absenceof reactants was merely a desorption from Pd/TiO2 that was inadsorption/desorption equilibrium with adsorbed water mole-cules, as confirmed by FTIR [21].

In the presence of 0.28 MPa of H2 only, no C-containing prod-ucts were detected above he detection limit (Table 4n00 00 0 andFig. 8), demonstrating that CO2 was the only C source in this study.Conversely, in the presence of 0.12 kPa of CO2 only, methanol wasproduced very slowly (0.008 lmol h�1 gcat�1; Table 4n00 00 00 and Fig. 8).As the rate was negligible compared with the methane formationrate in the presence of CO2 + H2 (38 lmol h�1 gcat�1), hydrogen wasthe predominant reductant for CO2 compared with adsorbed wateror surface hydroxy in this study.

In the absence of light at 298 K (the initial temperature 290 K inthe cell increased to 298 K after 10 min of UV–visible light irradia-tion; Table 4n00 0), thermal methane formation proceeded as in Eq.(11), but it was only 7.9% of that under UV–visible light irradiation

Fig. 7. The dependence of formation rates for CH4, CO, methanol, and water on totalreaction pressure using a Pd/TiO2 photocatalyst, where PCO2 =PH2 ¼ 3=7.

Fig. 8. Contour diagram for the total photoformation rates of C-containing productsfrom CO2 and H2 using a Pd/TiO2 photocatalyst as a function of the partial pressuresof H2 and CO2.

S. Kawamura et al. / Journal of Catalysis 345 (2017) 39–52 49

(Table 4n), demonstrating that most of the CO2 conversion fromCO2 and H2 in Table 4 proceeded photocatalytically.

To verify the thermal effects, blank tests in the absence of lightwere also carried out at 353 K (Table 4n00 00). The methane formationrate decreased by a factor of 0.43 times comparedwith that at 298 K.Moreover, 73% of the water formed among 9.8 lmol h�1 gcat�1

(Table 4n00 00) was not stoichiometric according to Eq. (11), but wasmerely simple desorption of adsorbed water from the TiO2 surfaceat 353 K.

4. Discussion

4.1. Energetics of the photoconversion of CO2 and H2

The dependence of CO2 photoconversion using H2 on the typeand amount of photocatalyst and the reaction pressure was stud-ied. For 10–100 mg of photocatalyst in 0.40 MPa of CO2 and H2,the activity order Eqs. (9) and (14) for formation of all C-containing compounds (methane, CO, and methanol) can be sum-marized as

Pd or Cu on TiO2 >> Ag=ZrO2 � g� C3N4

> Ag=Zn3Ga� LDH � BiOCl: ð15Þ

This order was not correlated to the Eg values of the metal oxide orhydroxide semiconductors (3.1–3.2, 5.1, 2.9, 5.2, and 3.6 eV; Table 1)and was almost in the opposite direction to the CB minimum poten-tials for the semiconductors (�0.1, �1.0, �1.3, �1.3, and �0.6 V atpH 0) [8,23,33,45] because more negative voltages are advanta-geous for CO2 reduction based on equilibrium control (Scheme 1B).Thus, the (excited) energy level of metal nanoparticles (Pd, Cu, orAg) for the electron trap and the efficiency of electron injection intoCO2-derived species were the major factors that determined thereaction order of Eq. (15), as discussed below.

Electron traps that decrease the recombination of photogener-ated charges were proposed for metal-doped TiO2 [11]. The workfunctions (WFs) for Pd, Cu, and Ag are 5.22–5.6, 4.48–5.10, and4.52–4.74 eV versus the vacuum level [66], corresponding topotentials of 0.78–1.16, 0.04–0.66, and 0.08–0.30 V versus SHE(pH 0), respectively, which are slightly positive compared withthe CB of TiO2 (�0.1 V; Scheme 1B). The OV sites for BiOCl arereported to exist at �0 V at pH 2 [23,67]. Thus, the potentials forelectron trapping levels for Pd/TiO2, Cu/TiO2, Ag/ZrO2, g-C3N4, Ag/Zn3Ga-LDH, and BiOCl are 0.78–1.16, 0.04–0.66, 0.08–0.30, �1.3,0.08–0.30, and �0 V, respectively. Except for g-C3N4, the electrontrapping level was very close to or a little more positive than thereduction potentials of methane (0.17 V at pH 0) or CO (0.30 or�0.12 V at pH 0) [14,68]. In general, the electron trapping levelbecame progressively more positive as the CO2 photoformationrates increased (Eq. (15)). Thus, the probability of direct electrontransfer from the trapping levels to CO2-derived species seems low.

Xie et al. reported a similar correlation between the potential ofelectron trapping sites and the photoformation rates of CO2 tomethane and CO using noble-metal-supported TiO2, which fol-lowed the order of the metal WFs: Pt > Pd > Au > Rh > Ag [11].Thus, we propose that the electron transfer to CO2-derived specieswas a combination of excited electron traps at Pd from TiO2 fol-lowed by transfer of hot/excited electrons produced by SPR toCO2-derived species (Scheme 1B). Thus the proposed reactionmechanism in the energetic diagram was in accord with the wave-length dependence of irradiation for Pd/TiO2 (Table 4n, n0, n00). Thesummation of rates under UV light and under visible light did notaccount for the rate under UV–visible light, implying a synergeticeffect, e.g., of UV for the band-gap excitation of TiO2 and of visiblelight for the hot electrons by SPR. The SPR peak(s) were centered at3.1–1.5, 2.8, 2.9, and 3.1 eV for Pd/TiO2, Cu/TiO2, Ag/ZrO2, and Ag/Zn3Ga-LDH (Table 1a, d, e, f), respectively. Thus, the levels of hot/excited electrons due to SPR were estimated to be �2.3 to �0.3,�2.8 to �2.1, �2.8 to �2.6, �1.3, �3.0 to �2.8, and �0 V for Pd/TiO2, Cu/TiO2, Ag/ZrO2, g-C3N4, Ag/Zn3Ga-LDH, and BiOCl, respec-tively, if we assume that the Fermi level for metal nanoparticlesdoes not shift.

In summary, as the levels of hot/excited electrons became morenegative, the catalysts became more active, except for Ag/ZrO2 andAg/Zn3Ga-LDH, probably due to lower charge separation efficiencyfor intrinsic semiconductor (ZrO2) [69] or hydroxide (Zn3Ga-LDH)[70,71]. Essentially, the hot/excited electrons could be transferredeasily to CO2-derived species based on the energy difference, andthen the positive charge remaining in the metal nanoparticlesrecombined with electrons that diffused from the semiconductors(Scheme 1B). However, hot/excited electron transfer by SPR wasnot enough to complete the reactions (11)–(13), as confirmed byquite poor activity in the photocatalytic test irradiated by light ofk > 420 nm (Table 4n00). Thus, oxidation site of H2 to protons shouldbe on TiO2, close to the interface with Pd, but not on Pd(Scheme 1A).

Scheme 1. Proposed reaction mechanism of CO2 photoconversion using H2 (A) andthe corresponding energetic diagram (B; pH 0).

50 S. Kawamura et al. / Journal of Catalysis 345 (2017) 39–52

Ag/Zn3Ga-LDH and Cu phthalocyanine(CuPc)/Zn3Ga-LDH alsocollected electrons due to SPR of Ag [28] or HOMO–LUMO excita-tion of the CuPc and BG excitation of the LDH [6,28]. Zhang et al.proposed a similar electron transfer to CO2 using TiO2 nanofibersdoped with Au/Pt [72]. The efficiency of trapping electrons in metalnanoparticles that had diffused from the semiconductor was sug-gested to control the activity order in Eq. (15). Wang et al. pro-posed a similar CO2 photoconversion mechanism using small Ptnanoparticles on approximately 2 lm of single-crystal TiO2 filmprepared by metal sputtering for 5–60 s [55]. Ultimately, small Ptnanoparticles approach the WF level of �2.128 V for Pt1 and wouldreduce CO2 directly. On the other hand, similar electron transfersfrom CB or OV sites to CO2 resulting in methane and/or CO forma-tion were already reported for g-C3N4 [73] and BiOCl [53,74].

The total formation rates of all C-containing compounds using100 mg of photocatalyst powder (0.28–0.025 lmol-C h�1;Table 4c–h) were clearly lower than those using 10 mg of photo-catalyst film (0.38–0.030 lmol-C h�1; Table 4i–k, n, v, w). Plausiblereasons for this are more effective absorption of UV–visible light incomparison to reflectance/diffraction (Fig. 2c, d) and possiblypreferable orientation of TiO2 crystallites for the 10 mg films com-pared with the 100 mg powders.

In summary, photoconversion rates of CO2 were controlled bythe energy of hot/excited electrons produced by SPR in metals,the CB minimum of g-C3N4, and OV sites in BiOCl. Cu or Pd onTiO2, especially as 10 mg films, was the most effective for photo-conversion of CO2 and H2.

4.2. Reaction mechanism for photoconversion of CO2 and H2 using themost active Pd/TiO2 catalyst

The dependence of CO2 conversion rates for the most active Pd/TiO2 photocatalyst on the partial pressures of CO2 and H2 wasspecific; the top of the volcano was searched for by following themethane formation rates along four straight lines followingPCO2 ¼ ð3=7ÞPH2 , PCO2 ¼ PH2 , PH2 = 0.28 MPa, and PCO2 = 0.12 MPa

and a steep volcano-like contour diagram could be drawn(Fig. 8). The rates reached a maximum at PCO2 = 0.12 MPa andPH2 = 0.28 MPa. In the zone PCO2 6 0.12 MPa and PH2 6 0.28 MPa,the reaction order for methane formation was 1.53 and the depen-dence was primarily determined by PH2 , suggesting that the Pd sur-face was preferentially covered with H and the reaction ordershould be mostly dependent on PH2 . In the partial-pressure zoneat 0.30–0.40 MPa, the mass balance of Eq. (11) was preserved(Table 4m, n).

In the proposed mechanism (Scheme 1A), hydrogen dissociatesand is oxidized by holes at the interface between Pd and the TiO2

surface. The protons that form progressively combine with CO2-derived species and electrons due to the SPR effects of Pd (Sche-me 1B), finally forming methane and water. If the protons do notencounter CO2-derived species, they can react with lattice oxygenat the interface between Pd and the TiO2 surface to form water andOV sites.

Conversely, in the zone PCO2 P 0.12 MPa and PH2 P 0.28 MPa,exclusive methane formation rates progressively dropped (Fig. 8),and deviations from the mass balance of Eq. (11) became more sig-nificant (Table 4o–q and Fig. 7). These trends were not due to thechange of Pd valence states. The Pd sites should be exclusivelymetallic or even the OV sites at the interface with TiO2 surfaceincreased after the CO2 photoconversion tests, as confirmed byPd K-edge XAFS (Table 2A–d, e and Fig. 6A-d, e, e0 and B-d, e).One of the reasons for the decrease in activity in this pressure zoneis that the Pd surface was predominantly covered with H, inhibit-ing the approach of CO2 (Scheme 1A). This difference is supportedby the difference in binding energies of CO2 (�0.33 eV) and H(�2.95 eV) on the Pd (111) surface [54]. Once the adsorption ofCO2 onto Pd sites became difficult, the conversion of CO2 tomethane and O2 decreased (Fig. 7), and intermediates and/or waterwould remain on the TiO2 surface, e.g., on OV sites formed duringpretreatment.

In the proposed reaction mechanism, the reaction of CO2-derived species, protons, and SPR electrons was plausible, but wesuspect that the thermally dissociated H species from H2 on thePd surface at 298 K would also combine with CO2-derived species.In fact, 7.9% of the methane formed under UV–visible illumination(Table 4n) could be formed under 0.40 MPa of CO2 and H2 after 5 hin the absence of light at 298 K (Table 4n00 0). Hence, dissociated Hfrom H2 on the Pd surface would be incorporated as a minor routeamong the multiple reduction steps involved in converting CO2 tomethane. Overall, the CO2 conversion basically proceededphotocatalytically.

Recently, CO2 activation at OV sites on TiO2 has often been sug-gested experimentally [10–13] and theoretically [75,76], in mostcases using water as a reductant via

CO2 þ OV ! COþ OðsurfaceÞ ð16Þand/or

CO2 þ Ti3þ ! Ti4þ � CO�2 ; ð17Þ

CO�2 þ Hþ þ e� ! COþ OH�: ð18Þ

The rate of CO2 photoconversion into methane using Pd/TiO2

(38 lmol h�1 gcat�1; Table 4n) was higher than those into methaneand methanol using undoped TiO2 (4.0 lmol h�1 gcat�1; Table 4w) orreported in the literature (0.15–17 lmol h�1 gcat�1) [11,12,56,72,77–80] by 1–2 orders of magnitude, suggesting the importance of Pdsites for CO2 reduction. However, CO2 activation via Eq. (16) and/or (17) + (18) is, at most, only partial in this study (Scheme 1A).

The negligible activity using Pd/TiO2 irradiated by visible lightonly (0.009 lmol-CH3OH h�1 gcat�1 Table 4n00), which was even lowerthan that in the absence of light at 298 K (3.0 lmol-CH4 h�1 gcat�1

S. Kawamura et al. / Journal of Catalysis 345 (2017) 39–52 51

Table 4n00 0), needs attention. Under the irradiation of visible lightonly, the formation of protons from H2 should be negligible andhot electrons by SPR effect diffused in Pd nanoparticles (Sche-me 1A). The minor reaction path of CO2 with thermally formed Hon the Pd surface would be suppressed by the electrons becausethe reaction of CO2 with negatively charged H should be difficultcompared with the coupling of CO2 with H+ or H.

Finally, the possibility of thermal CO2 conversion is considered.This study employed a similar light source to Ahmed et al. [8] andYoshida et al. [65,81], in which the maximum temperature reached313 K. The pressure decrease due to methane formation via Eq.(11) at a rate of 38 lmol h�1 gcat�1 under 0.40 MPa of CO2 and H2

(Table 4n) is as small as 93 Pa. As the critical pressures for CO2

and H2 are high (7.38 and 1.30 MPa, respectively) compared withthe reaction pressures (0.12 and 0.28 MPa, respectively), the reac-tion gas can be approximated as an ideal gas. The reaction pressurebefore the irradiation of UV–visible light (0.400 MPa) graduallyincreased to 0.412 MPa after approximately 10 min of irradiationand remained constant for 5 h. After the UV–visible light wasturned off, the pressure decreased to 0.400 MPa in 30 min. Thus,the initial temperature (290 K) reached thermal equilibrium at298 K during the test. Under the same reaction conditions butusing a white TiO2 photocatalyst (Table 4w), the thermal-equilibrium pressure was 0.406 MPa, corresponding to a reactiontemperature of 294 K. Furthermore, partial heating of the catalystalso needs attention. The heat of reaction for Eq. (11)(165 kJ mol�1) should elevate the Pd/TiO2 catalyst by 45 K (if themolar heat capacity at constant pressure is assumed to be55.31 J mol�1 K�1 for anatase-type TiO2) at a maximum, althoughpart of the heat should diffuse to the Pyrex glass substrate beneaththe catalyst film (Fig. 1C). The blank test in the absence of UV–vis-ible light using Pd/TiO2 in CO2 and H2 at 353 K, which is 18 Khigher than the theoretical maximum temperature for the Pd/TiO2 catalyst, produced only 3.4% methane compared with theUV–visible light irradiated test (Table 4n, n00 00), reconfirming thatUV–visible light was indispensable for the conversion of CO2 usingH2 in this study.

In summary, no reactions proceeded in the absence of reactants(Table 4n00 00 00 0) or catalyst (Table 4y). 7.9–3.4% of the methane wasproduced in the absence of UV–visible irradiation (Table 4n00 0,n00 00). Thermal processes accounted for a minor fraction of reaction(11), but their rate decreased to 0.43 times the rate at 298 K whenthe reaction temperature was elevated to 353 K, due to the com-plex adsorption/reaction steps for reaction (11).

5. Conclusions

Photocatalytic CO2 conversion was investigated up to 0.80 MPausing H2 and 10–100 mg of Pd/TiO2, Cu/TiO2, Ag/ZrO2, g-C3N4, Ag/Zn3Ga-LDH, and BiOCl. The catalytic activities were in the order Pdor Cu on TiO2 � Ag/ZrO2 � g-C3N4 > Ag/Zn3Ga-LDH � BiOCl. A two-step electron excitation mechanism was strongly suggested fordoped metal nanoparticles, in which hot/excited electrons arisingfrom SPR effects were estimated to exist at �2.8 to �0.3 V vs.SHE, which is more negative than the potentials for the reductionof CO2 to methane or CO (�0.12 to 0.30 V), and the remaining pos-itive charge should combine with excited electrons that have dif-fused from the semiconductor. Therefore, as the levels of hot/excited electrons became more negative, the catalysts basicallybecame more active except, for intrinsic semiconductor (ZrO2)and hydroxide (LDH) [82]. For the oxidation reaction, H2 was sug-gested to couple with photogenerated holes to form protons at theinterface between Pd and the TiO2 surface. The protons wouldcombine progressively with CO2-derived species and electrons atthe Pd to form methane and water, and would also react with lat-

tice O to form OV sites at the interface between Pd and TiO2, as con-firmed by Pd K-edge EXAFS.

Based on blank tests in the absence of UV–visible light at 298–353 K, thermal CO2 conversion under H2 at 0.40 MPa was 7.9–3.4%of that under UV–visible light irradiation. Therefore, the CO2 con-version in this study is almost exclusively photocatalytic, but ther-mal H adsorption on Pd from H2 is also plausible. H was preferablyadsorbed on Pd, leading to the CO2 photoconversion efficiencydepending exclusively on PH2 . The methane formation rate reacheda maximum (38 lmol h�1 gcat�1) at PCO2 = 0.12 MPa andPH2 = 0.28 MPa, but the rate dropped at higher pressures due tohindrance of CO2 adsorption by H.

Acknowledgments

The authors are grateful for financial support from Grant-in-Aidfor Scientific Research C (26410204) from the Japan Society for thePromotion of Science and the Leading Research Promotion Program(2015) from the Institute for Global Prominent Research, ChibaUniversity. X-ray absorption experiments were conducted underthe approval of the Photon Factory Proposal Review Committee(2015G586 and 2014G631). The authors thank Sakino Oshiro forsample preparation and kinetic measurements. The authors thankEnago (www.enago.jp) for the English language review.

Appendix A. Supplementary material

Supplementary data associated with this article can be found, inthe online version, at http://dx.doi.org/10.1016/j.jcat.2016.10.024.

References

[1] S.N. Habisreutinger, L. Schmidt-Mende, J.K. Stolarczyk, Angew. Chem. Int. Ed.52 (2013) 7372.

[2] Y. Izumi, Advances in CO2 capture, sequestration, and conversion, in: F. Jin, L.-N. He, Y.H. Hu (Eds.), ACS Books, vol. 1194, 2015 (p. 1, Chapter. 1).

[3] Y. Izumi, Coord. Chem. Rev. 257 (2013) 171.[4] C.C. Yang, Y.H. Yu, B. van der Linden, J.C.S. Wu, G. Mul, J. Am. Chem. Soc. 132

(2010) 8398.[5] T. Yui, A. Kan, C. Saitoh, K. Koide, T. Ibusuki, O. Ishitani, A.C.S. Appl, Mater.

Interf. 3 (2011) 2594.[6] S. Kawamura, N. Ahmed, G. Carja, Y. Izumi, Oil Gas Sci. Technol. 70 (2015) 853.[7] N. Ahmed, M. Morikawa, Y. Izumi, Catal. Today 185 (2012) 263.[8] N. Ahmed, Y. Shibata, T. Taniguchi, Y. Izumi, J. Catal. 279 (2011) 123.[9] D. Voet, J.G. Voet, Biochemistry, second ed., J. Wiley & Sons, New York, 1995, p.

639.[10] S�. Neaţu, J.A. Maciá-Agulló, P. Concepción, H. Garcia, J. Am. Chem. Soc. 136

(2014) 15969.[11] S. Xie, Y. Wang, Q. Zhang, W. Deng, Y. Wang, ACS Catal. 4 (2014) 3644.[12] L. Liu, H. Zhao, J.M. Andino, Y. Li, ACS Catal. 2 (2012) 1817.[13] L. Liu, Y. Jiang, H. Zhao, J. Chen, J. Cheng, K. Yang, Y. Li, ACS Catal. 6 (2016) 1097.[14] F. Galli, M. Compagnoni, D. Vitali, C. Pirola, C.L. Bianchi, A. Villa, L. Prati, I.

Rosetti, Appl. Catal. B 200 (2017) 386.[15] I. Rossetti, A. Villa, M. Compagnoni, L. Prati, G. Ramis, C. Pirola, C.L. Bianchi, W.

Wang, D. Wang, Catal. Sci. Technol. 5 (2015) 4481.[16] I. Rossetti, A. Villa, C. Pirola, L. Prati, G. Ramis, RSC Adv. 4 (2014) 28883.[17] R. Li, J. Hu, M. Deng, H. Wang, X. Wang, Y. Hu, H.L. Jiang, J. Jiang, Q. Zhang, Y.

Xie, Y. Xiong, Adv. Mater. 26 (2014) 4783.[18] Y. He, Y. Wang, L. Zhang, B. Teng, M. Fan, Appl. Catal. B 168 (169) (2015) 1.[19] M.A. Asi, L. Zhu, C. He, V.K. Sharma, D. Shu, S. Li, J. Yang, Y. Xiong, Catal. Today

216 (2013) 268.[20] F. Sastre, A.V. Puga, L. Liu, A. Corma, H. García, J. Am. Chem. Soc. 136 (2014)

6798.[21] H. Zhang, S. Kawamura, M. Tamba, T. Kojima, M. Yoshiba, Y. Izumi, in

preparation.[22] J. Liu, T. Zhang, Z. Wang, G. Dawson, W. Chen, J. Mater. Chem. 21 (2011) 14398.[23] Y. Fujishima, S. Okamoto, M. Yoshiba, T. Itoi, S. Kawamura, Y. Yoshida, Y.

Ogura, Y. Izumi, J. Mater. Chem. A 3 (2015) 8389.[24] A. Sandoval, C. Louis, R. Zanella, Appl. Catal. B 140 (141) (2013) 363.[25] J.A. Bearden, Rev. Mod. Phys. 39 (1967) 78.[26] G. Zschornack, Handbook of X-ray Data, Springer, Berlin, Heidelberg, 2007, p.

233.[27] F. Wooten, Optical Properties of Solids, Academic Press, New York, 1972, p. 14.[28] S. Kawamura, M.C. Puscasu, Y. Yoshida, Y. Izumi, G. Carja, Appl. Catal. A 504

(2015) 238.

52 S. Kawamura et al. / Journal of Catalysis 345 (2017) 39–52

[29] Y. Ogura, S. Okamoto, T. Itoi, Y. Fujishima, Y. Yoshida, Y. Izumi, Chem.Commun. 50 (2014) 3067.

[30] F.W. Lytle, R.B. Greegor, D.R. Sandstrom, E.C. Marques, J. Wong, C.L. Spiro, G.P.Huffman, F.E. Huggins, Nuclear Instrum. Methods Phys. Res. 226 (1984) 542.

[31] Y. Izumi, H. Nagamori, F. Kiyotaki, D. Masih, T. Minato, E. Roisin, J.-P. Candy, H.Tanida, T. Uruga, Anal. Chem. 77 (2005) 6969.

[32] Y. Izumi, K. Konishi, D. Obaid, T. Miyajima, H. Yoshitake, Anal. Chem. 79 (2007)6933.

[33] Y. Izumi, T. Itoi, S. Peng, K. Oka, Y. Shibata, J. Phys. Chem. C 113 (2009) 6706.[34] Y. Izumi, Y. Shibata, Chem. Lett. 38 (2009) 912.[35] M. Vaarkamp, H. Linders, D. Koningsberger, XDAP Version 2.2.7, XAFS Services

International, Woudenberg, The Netherlands, 2006.[36] W.M. Haynes, D.R. Lide, T.J. Bruno, CRC Handbook of Chemistry and Physics,

96th ed., CRC Press, Boca Raton, USA, 2015.[37] A.G. Christy, S.M. Clark, Phys. Rev. B 52 (1995) 9259.[38] B.G. Hyde, S. Andersson, Inorganic Crystal Structures, John Wiley & Sons, New

York, 1989.[39] D.T. Cromer, K. Herrington, J. Am. Chem. Soc. 77 (1955) 4708.[40] http://www.tokina.co.jp/io-filters/filters/sharp-cut-filters.html.[41] http://www.tokina.co.jp/io-filters/filters/ut-va-filter.html.[42] M. Morikawa, Y. Ogura, N. Ahmed, S. Kawamura, G. Mikami, S. Okamoto, Y.

Izumi, Catal. Sci. Technol. 4 (2014) 1644.[43] K.-H. Choi, N. Duraisamy, M.N. Awais, N.M. Muhammad, H.C. Kim, J. Jo, Mater.

Sci. Semiconduct. Proc. 16 (2013) 1285.[44] S.N. Tkachev, M.H. Manghnani, A. Niilisk, J. Aarik, H. Mändar, J. Mater. Sci. 40

(2005) 4293.[45] Y. Fu, T. Huang, L. Zhang, J. Zhu, X. Wang, Nanoscale 7 (2015) 13723.[46] K.H. Leong, H.Y. Chu, S. Ibrahim, P. Saravanan, Beilstein J. Nanotechnol. 6

(2015) 428.[47] K. Thamaphat, P. Limsuwan, B. Ngotawornchai, Kasetsart J. (Nat. Sci.) 42 (2008)

357.[48] C. Clavero, Nat. Photon. 8 (2014) 95.[49] Y. Izumi, H. Kurakata, K. Aika, J. Catal. 175 (1998) 236.[50] K. Aika, J. Kubota, Y. Kadowaki, Y. Niwa, Y. Izumi, Appl. Surf. Sci. 121–122

(1997) 488.[51] Y. Izumi, K. Aika, J. Phys. Chem. 99 (1995) 10336.[52] Y. Izumi, Y. Iwasawa, ChemTech 24 (1994) 16.[53] B.J. Kip, F.B.M. Duivenvoorden, D.C. Koningsberger, R. Prins, J. Catal. 105 (1987)

26.[54] L. Dietz, S. Piccinin, M. Maestri, J. Phys. Chem. C 119 (2015) 4959.

[55] W.N. Wang, W.J. An, B. Ramalingam, S. Mukherjee, D.M. Niedzwiedzki, S.Gangopadhyay, P. Biswas, J. Am. Chem. Soc. 134 (2012) 11276.

[56] J. Yu, J. Low, W. Xiao, P. Zhou, M. Jaroniec, J. Am. Chem. Soc. 136 (2014) 8839.[57] L. Zhang, W. Wang, D. Jiang, E. Gao, S. Sun, Nano Res. 8 (2015) 821.[58] Y. Iwasawa, Adv. Catal. 35 (1987) 187.[59] B.B. Hallac, K. Keyvanloo, J.D. Hedengren, W.C. Hecker, M.D. Argyle, Chem. Eng.

J. 263 (2015) 268.[60] G. Vella, G.E. Imoberdorf, A. Sclafani, A.E. Cassano, O.M. Alfano, L. Rizzuti, Appl.

Catal. B 96 (2010) 399.[61] X. Ouyang, L. Bednarova, R.S. Besser, P. Ho, AIChE J. 51 (2005) 1758.[62] F. Humblot, D. Didillon, F. Lepeltier, J.P. Candy, J. Corker, O. Clause, F. Bayard, J.

M. Basset, J. Am. Chem. Soc. 120 (1998) 137.[63] D. Masih, H. Yoshitake, Y. Izumi, Appl. Catal. A 325 (2007) 276.[64] G. Mikami, F. Grosu, S. Kawamura, Y. Yoshida, G. Carja, Y. Izumi, Appl. Catal. B

199 (2016) 260.[65] Y. Yoshida, Y. Mitani, T. Itoi, Y. Izumi, J. Catal. 287 (2012) 190.[66] W.H. Haynes (Ed.), CRC Handbook of Chemistry and Physics, CRC Press, New

York, USA, 2015 (section 12–122).[67] L. Ye, K. Deng, F. Xu, L. Tian, T. Peng, L. Zan, Phys. Chem. Chem. Phys. 14 (2012)

82.[68] C. Wang, R.L. Thomson, J. Baltrus, C. Matranga, J. Phys. Chem. Lett. 1 (2010) 48.[69] J. Zhang, L. Li, S. Wang, T. Huang, Y. Hao, Y. Qi, RSC Adv. 6 (2016) 13991.[70] X. Liu, X. Zhao, Y. Zhu, F. Zhang, Appl. Catal. B 140–141 (2013) 241.[71] M. Morikawa, N. Ahmed, Y. Yoshida, Y. Izumi, Appl. Catal. B 144 (2014) 561.[72] Z. Zhang, Z. Wang, S.W. Cao, C. Xue, J. Phys. Chem. C 117 (2013) 25939.[73] W.-J. Ong, L.-L. Tan, Y.H. Ng, S.-T. Yong, S.-P. Chai, Chem. Rev. 116 (2016) 7159.[74] J. Jin, Y. Wang, T. He, RSC Adv. 5 (2015) 100244.[75] Y. Ji, Y. Luo, ACS Catal. 6 (2016) 2018.[76] D. Lee, Y. Kanai, J. Am. Chem. Soc. 134 (2012) 20266.[77] N.M. Dimitrijevic, B.K. Vijayan, O.G. Poluektov, T. Rajh, K.A. Gray, H. He, P.

Zapol, J. Am. Chem. Soc. 133 (2011) 3964.[78] H. Zhao, L. Liu, J.M. Andino, Y. Li, J. Mater. Chem. A 1 (2013) 8209.[79] L. Ye, J. Mao, T. Peng, L. Zan, Y. Zhang, Phys. Chem. Chem. Phys. 16 (2014)

15675.[80] B.S. Kwak, K. Vignesh, N.K. Park, H.J. Ryu, J.I. Baek, M. Kang, Fuel 143 (2015)

570.[81] Y. Yoshida, T. Itoi, Y. Izumi, J. Phys. Chem. C 119 (2015) 21585.[82] M. Miyano, H. Zhang, M. Yoshiba, Y. Izumi, Energy Technol., http://dx.doi.org/

10.1002/ente.201600578.